IMAGE PROCESSING OF CONVENTIONAL X-RAY IMAGES
Dr. Khalil I. Jassam,
Researcher, The Institute of Islamic Medicine for Education and Research, Panama City, FL
and Visiting Professor, Department of Surveying Engineering, University of Maine, Orono, ME
USA, Commission No: VII
ABSTRACT:
The goal of this paper is to outline the procedure of obtaining sharper and more visible images from a
rejected X-ray. This process will improve X-ray image quality and produce image data that is more
effectively displayed for a later visual imaging diagnosis. Image processing enhances image contrast
thus increasing image visibility, helping both physicians and radiologists to make more accurate
diagnoses and to decrease the need to retake X-rays. This in turn reduces the risk of radiation
exposure and increases economical benefits by lessening the number of rejected X-rays. Different
spectral and spatial enhancement techniques were used both in the spatial and frequency domain. The
obtained X-ray images are sharper, more visible and recognizable, and provide much more
information.
Key Words: X-ray Imaging, Image Processing, Medical Imaging.
INTRODUCTION
BACKGROUND
The discovery of X-rays revolutionized the diagnosis
procedure, and its importance can not be overemphasized.
Radiographie quality refers to both image visibility and
recognizability. The visibility of the image is best when its
density is sufficient, its noise is minimal, and its contrast is
maximum. It is most recognizable when its geometry is
maintained, which takes place when sharpness is
maximized and image distortion and magnification are
minimized. Several factors affect the image quality, some
of which are the focal-spot size, milliampere-seconds,
kilovoltage, field size limitation, patient status, contrast
media, and film quality. Over the years the optimum
parameters for a specific examination have been empirically
determined by a large number of practitioners.
X-rays were discovered in 1895 by the German physicist
Roentgen and were so named because the1r nature was
unknown at the time. Unlike ordinary light, these rays are
invisible, but they travel in straight lines and affect
photographic film in the same way as light. On the other .
hand, they were much more penetrating than light and
could easily pass through the human body, wood and other
"opaque" objects. We know today that X-rays are
electromagnetic radiation of exacdy the same nature as light
but of very much shorter wavelength. The unit of
measurement in the X-ray region is the angstrom (A), equal
to 1O-8cm . X-rays, used in diffraction, have wavelengths
lying approximately between the range of 0.5 - 2.5 A,
where the wavelength of visible light is on the order of
6000 A. X-rays therefore occupy the region between
gamma and ultraviolet rays in the electromagnetic spectrum
(figure 1).
Tremendous efforts have been invested in upgrading X-ray
image quality. A variety of techniques were developed,
which were mainly concerned with hardware improvement,
but their effects were limited. In the last decade computed
tomography (CT) was developed. This system represents
the state of the art in modern X -ray imaging. The CT
system has major advantages as weIl as disadvantages. It
maximizes both image visibility and recognizability, and it
has a better resolution when compared to the conventional
X-ray. The main disadvantage of the CT system is that it is
too expensive to buy and operate, and only major clinics
can afford it. In addition, it is less safe due to higher
radiation levels and more expensive to the patiants. For
these reasons, the need for the conventional X-ray will
continue for the next decade.
The method employed to produce X-rays is essentially the
same as that used at the time of its discovery. A beam of
electrons accelerated by high voltage to a velocity
approaching the speed of light is rapidly decelerated upon
colliding with a heavy metal target. In the process of
slowing down, X-ray photons are emitted; the emitted Xray is then directed to the human body. The number of Xrays that interact with the patient depend upon the thickness
and the composition of the various tissues. Diagnostic Xrays interact primarily by the photoelectric and Compton
processes. Photoelectric interactions are the most
important for image formation because of the strong
dependence of the photoelectric effect on the atomic
composition of the absorber and the absence of long-range
secondary radiations. Compton interactions are generally
detrimental in that the likelihood of their occurrence
depends mainly on tissue density, and the scattered X-rays
produced in Compton collisions have a high probability of
escaping from the patient and crossing the image plane.
This paper expands the use of image processing techniques
to improve the quality of conventional X-ray images
without hardware modification. The only additional
hardware needed is a digitizing device and a personal
computer.
245
may be expressed in terms of variations in photon fluency,
variations in energy fluency or variations in exposure.
00
S
Wavelengh (J.lnJ )
X-ray images are formed in a manner similar to the regular
black and white pictures. Body parts which have higher
resistances to X-ray penetration ( bones) result in less light
reaching the film, and consequently brighter image on the
X-ray transparency which is nothing more than a negative
image. On the other hand, soft tissues have less resistance
to X-rays, so more X-rays pass through them, resulting in
more radiation reaching the film and producing darker tone.
In most cases the bones are the brightest and gases are the
darkest.
Television
and radio
s
.5
(j
~
Near- IR
0..
C/J
.g
Red
~
Green
Sb
C'($
Blue
Ei
.9
(j
uv
~
~
§.
'I?
S
.9
S
Ei
0
..;::
~
~
M
~
a
"t~
a
u::
...c::
2
1"
so='~
eJ)
1-0
's
~
C/J
>-.
e
Responses
~
Figure 2. Formation of radiological image
V">
's
\0
's
Xrays
Gammarays
THE REJECTED X-RA Y PICTURES
Cosmicrays
Rejected X-ray pictures are the ones which do not satisfy
the initial intended purpose. They are usually thrown away
or destroyed, and the radiographs have to be retaken. Xray images may be rejected for more than one reason, for
example: over or under exposure, patient movement, poor
film development, and other causes which result in poor
contrast. (Figures 3 and 4 show examples of rejected Xray images.)
~
Q)
~
~
Because their directions are unrelated to the position of the
focal spot, these scattered X-rays do not carry any useful
information about the patient and serve only to reduce Xray contrast. Unfortunately, the interaction of diagnostic
X-rays with soft tissue is mainly by the Compton process,
and specific stratagems must be employed to prevent
"scatter" from reaching the imaging device. The X-ray
image is determined by the intensity distribution in the Xray beam as it emerges from the patient. The quality, i.e.
visibility and recognizability of the X-ray image, depends
upon the focal-spot size, the incident X-ray spectrum, and
the composition of the patient. Over many years the
optimum parameters for a specific examination (e.g. X-ray
tube potential, beam filtration, exposure time, infection of
contrast media) have been empirically determined by a large
number of practitioners.
Rejected images account for 15 to 20 percent ofthe X-rays
taken per year. This in turn costs public and private
hospitals and clinics millions of dollars and exposes
patients to unnecessary radiation which is a major public
concern. They also slow down the diagnostic process and
increase the cost which is passed on to the patients.
Presently, the only solution radiologists and physicians
offer is to destroy X-rays and take new ones. This has
been and probably will be, the trend for the next decade.
With the growing public concern of insurance cost, and
with increase in computer capabilities both in software and
hardware, the medical community have to find a better and
more reasonable solution.
At average diagnostic kilovoltage levels, about 5% or less
of the primary radiation traverses completely through the
patient's body, without interacting with any of the atoms in
the patient, and strikes the film. In addition, about 15% of
the primary radiation interacts with atoms resulting in the
production of the secondary photons which make it out of
the patient and strike the film. The remaining 80% of the
primary beam is totally absorbed within the patient. Figure
2 illustrates the attenuation of an X-ray beam by the various
tissues within the patient, resulting in a variation of
transmitted radiation. The pattern of transmitted radiation
246
Display
Figure 5. Thc Propose System
Results show that linear and non-linear contrast
enhancements increased the visual quality of the images
and in many cases the new image is no Ionger c1assified as
a rejected X-ray. Spatial enhancement has the effect of
bringing up the details of the image. Logarithmic and
exponential histogram equalization gave good results in
most cases. A combination of linear stretch and histogram
equalization also proved to be effective enhancement
Figure 3. A rejected over exposed X-ray
techniques. High boost filter with central kernel value less
than 10 gave the best result in most cases. High-frequency
emphasis filter also proved to be a powerful tool for spatial
enhancements. Different edge enhancement filters were
examined and directional and Sobel filter worked the best.
Figures 6, 7,8 and 9 show some of the obtained results.
These figures demonstrates that indeed a good part of the
rejected X-rays can be retrieved. This process not only
reduces number of rejected X-rays, but it also increases the
quality of both the rejected and the good X-rays. The new
obtained X-ray is sharper anel more visually interpretable.
The procedure will save both patient and physician time
and money and will be much safer than the present day
system.
Figure 4. A rejected under exposed X-ray
THE PROPOSED SYSTEM
Following is one possible solution to minimize the number
of the rejected X-rays. The heart of the solution is the
concept of image processing. All X-rays can be converted
to digital images by simple or advanced scanning
techniques. Once a digitized X-ray is obtained basic and
advanced image processing techniques can be applied to it.
Research shows that more than 90% of the rejected images
are due to over- or under-exposure or to poor contrast in
the obtained radiographs. Less than 10% of the X-rays are
rejected because of the patient moving. In these cases
image restoration and enhancements can be applied to
minimize ,md/or correct the problem.
The proposed system (figure 5) is nothing more than a
typical image processing unit connected to an X-ray
machine.
Figure 6. Exponential equalization for the
X-ray in figure 3
Six rejected X-ray pictures were selected and then were
digitized at different resolutions ranging from 50 to 300 11m
pixel size. Initial statistics were collected to select proper
restoration techniques. Spectral anel spatial image
enhancement techniques were appliecl to the digital image.
247
Figure 9-B. Exponential equalization for the X-ray in 9-A,
notice the lighter tone in the middle of the joint
Figure 7. Sobel filter for the X-ray in figure 3
Limitations and Practical Difficulties
This system would produce the expected result by
minimizing the number of rejected X-ray pictures.
However, several practicallimitations should be addressed:
Resolution
The current X-ray film capture data and provide excellent
resolution. Converting X-ray pictures from pictorial to
digital would require the use of certain type of scanner.
Currently there are many scanners in the market ranging
from several hundred dollars to hundreds of thousands of
dollars. Resolution and image quality obtained from these
scanners increases with the prices. A scanner with 300
DPI might cost Iess than a thousand dollars, but the quality
of the obtained image is not suitable for medical diagnosis.
On the other hand, the quality of images scanned with 1200
DPI would be much higher and provide reasonable
resolution for medical diagnosis, but it is much more
expensive
Figure 8. Simple histogram equalization of the
X-ray in figure 4
Independent of scanner, the screen resolution for most
commercially available monitors is 1024 X 1024.
Therefore, if an image is scanned with 1200 DPI, and
displayed on a 1024 X 1024 screen, the resolution will be
less.
Size of Data Involved
Independent of scanner and monitor types, the obtained
digitized image would require significant dynamic and
static storage space. For example, a 15 X 16 inch chest Xray digitized at 300 DPI would produce an image 4500 X
4800 pixels. This would need 22 megabytes of memory
space. High quality digital image (1200 DPI) would
require 345 megabytes. Any enhancement or restoration
operations would require additional space.
Figure 9-A. A rejected over exposed joint
Time
Currently it takes two minutes to obtain an X-ray image.
This time includes taking the picture and developing the
film. In the proposed system, the X-ray would be taken,
would be digitized, and then processed. This would
require additional time especially if a high quality image is
248
needed, which would make the system less efficient to the
existing one.
Ljungberg, M., Strand, S., 1990s. Scatter and attenuation
correction in SPECT using density maps and Monte Carlo
simulated scatter functions. The Journal of Nuc1ear
Medicine. 31 (9) : 1560-1567.
Replacement of the film with electronic detectors would
eliminate the digitization phase and increase the system
efficiency but also increase the cost.
Robb, R. A., Barillot, C., 1989s. Interactive display and
analysis of 3-D medical images. IEEE Transactions on
Medicallmaging. 8 (3) : 217-225.
Attitude
Many radiologists and physicians are resisting any attempts
to computerize conventional X-ray images. They view it as
an unnecessary development process and see the existing
system as the best that can be obtained with current
technology.
Sezan, M. I., Tekalp, A. M., Schaetzing, R., 1989j.
Automatic anatomically selective image enhancement in
digital chest radiography. IEEE Transactions on Medical
Imaging. 8 (2) :154-163.
An the above limitations will be reduced, and probably will
become insignificant, as development of computer
hardware and software advances. At that point, the
negative attitude of the medical community will also be
lessened and eventually digital X-ray would become a
reality.
Sherouse, G. W., Novins, K., Chaney, E. L., 1989.
Computation of digitally reconstructed radiographs for use
in radiotherapy treatment design. I. 1. Radiation Oncology
Biol Phys. 18 (3) : 651-658.
Sklarew, R. l, Bodmer, S. c., Pertschuk, L. P., 1989s.
Quantitative Imaging of Immunocytochemical (PAP)
estrogen receptor staining patterns in breast cancer
sections. Cytometry. 11: 359-378.
Conc1usion
Image processing techniques can be used to minimize
. significantly the number of rejected X -rays. Enhanced
image contrast and visibility will help both physicians and
radiologists make more accurate diagnosis. This in turn
will reduce the risk of radiation exposure and increase the
economical benefits. The enhanced X-ray images are
sharper, more visible and recognizable, and provide more
information.
Stone, J. L., Peterson, R. L., Wolf, 1. E., 1990n. Digital
imaging techniques in dermatology. Journal of the
American Academy ofDermatology. 23 (5) : 913-917.
Tahoces, P. G., Correa, J., Souto, M., Gonzalez, C.,
Gomez, L., Vidal, J. 1., 1991s. Enhancement of chest and
breast radiographs by automatic spatial filtering. IEEE
Transactions on Medical Imaging. 10 (3): 330-335.
References
Gonzalez, Rafael c., 1987. Digital Image Processing.
Addison-Wesley, Reading, Massachusetts, pp 61-201
Turner, D. A., Alcorn, F. S., Adler, Y. T., 1988. Nuc1ear
magnetic resonance in the diagnosis of breast cancer.
Radiologic Clinics of North America. 26 (3) : 673-684.
Fahimi, H., Macovski, A., 1989 m. Reducing the effects
of scattered photons in X-ray projection imaging. IEEE
Transactions on Medical Imaging. 8 (1) : 56-63.
Jain, Anil K., 1989. Fundamentals ofDigital Image
Processing. Prentice-Hall, New Jersey, pp. 431-471
Voillot, C., Gravier, M. and Ramaz, A., Chaix, J. M.,
1990m. The use of X-ray image processing to analyze the
z-direction distribution of fillers and pigments. Tappi
Journal. 191-193.
Kergosien, L. Y., 1991s. Generic sign systems in medical
imaging. IEEE Computer Graphics & Applications.
pp 46-65.
Wallis, J. W., Miller, T. R., 1990. Three dimensional
display in nuc1ear medicine and radiology. The Journal of
Nuc1ear Medicine. 32 (3) : 534-546.
King, P. L., Browning, R., Pianetta P. and Lindau I.,
Keenlyside M. and Knapp G., 1989j. Image processing of
multispectral X-ray photoelectron spectroscopy images. J.
Vac. Sei. Technol. 7 (6): 3301-3304.
Wallis, J. W., Miller, T. R., 1989d. Volume Rendering in
three dimensional display of SPECT images. The Journal
of Nuc1ear Medicine. 31 (8): 1421-1430.
Yokoi, S., Yasuda, T., Toriwaki, J., 1990d. A simulation
system for craniofacial surgeries based on 3D image
processing. IEEE Engineering in Medicine and Biology.
29-31.
Kreipke, D. L., Silver, D. I., Traver, R. D., Braunstein,
E. M., 1989m. Readability of cervical spine imaging:
digital versus film/screen radiographs. Computerized
Medical Imaging and Graphics, 14 (2): 119-125.
249